Development of an AI-driven drone system for precision forest health monitoring

Thesis Abstract

The deterioration of forest ecosystems due to climate change, pests, diseases, and deforestation necessitates the development of innovative monitoring tools that enhance early detection, improve management efficiency, and support sustainable forest stewardship. This study aims to develop a comprehensive AI-driven drone system tailored for precision forest health monitoring, integrating advanced sensor technologies and machine learning algorithms to facilitate real-time assessment of forest conditions. The specific objectives are to design and prototype an unmanned aerial vehicle (UAV) platform equipped with multispectral and LiDAR sensors, to develop and train machine learning models for identifying forest stress indicators, and to evaluate the system’s operational efficacy and accuracy across diverse forest types. The research adopts a mixed-methods approach within a quantitative-dominant paradigm, employing a quasi-experimental design to assess the system's performance. The population comprises forested regions within the temperate zone of the Pacific Northwest, encompassing 12 distinct forest plots selected based on variations in tree species, stand age, and disturbance history. A stratified random sampling technique was used to select 4 plots for detailed monitoring, with a sample size of 120 drone flights (10 flights per plot) conducted over a 6-month period to capture seasonal dynamics. Data collection involved deploying the drone system—equipped with multispectral and LiDAR sensors—to capture high-resolution imagery and point cloud data. Ground-truth data for model training and validation were obtained through on-site forest health assessments by certified arborists, including measurements of chlorophyll fluorescence, canopy discoloration, and pest infestation signs. The data analysis involved preprocessing of multispectral and LiDAR data using ENVI and LAStools software, followed by feature extraction to identify indicators relevant to forest stress. Supervised machine learning algorithms—including Random Forest, Support Vector Machine (SVM), and convolutional neural networks (CNN)—were employed to develop classifiers that distinguish healthy versus stressed forest conditions. Model performance was evaluated based on accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC-ROC). Spatial analysis using GIS tools assessed the spatial distribution of forest health indicators, while statistical analysis through regression analysis identified the significance of various spectral and structural variables in predicting forest stress. Theoretical frameworks underpinning this study include the Ecological Niche Theory and the Technology Acceptance Model, guiding the integration of environmental parameters and user adoption perspectives. Expected findings indicate that the AI-driven drone system can detect early signs of forest stress with an overall accuracy exceeding 85%, significantly outperforming manual survey methods in speed and spatial coverage. The machine learning models are anticipated to identify specific spectral and structural features strongly correlated with pest infestations, disease outbreaks, and drought stress, thereby enabling targeted intervention strategies. The study will contribute to existing knowledge by demonstrating a scalable, cost-effective approach to forest health monitoring that merges remote sensing, AI, and unmanned systems, filling notable gaps in the literature regarding operational deployment and model generalizability across contrasting forest ecosystems. The main conclusion highlights the potential of AI-powered drone systems to transform forest health surveillance into a proactive, precise, and efficient process, supporting policymakers and forest managers in decision-making. Recommendations include integrating the system into existing forest management frameworks, upgrading sensor capabilities to include hyperspectral imaging for enhanced diagnostics, and establishing standardized protocols for operational deployment. The study advocates for further research to explore longitudinal monitoring over longer timescales, customization of machine learning models to different forest types, and the development of user-friendly interfaces to facilitate wider adoption among forestry professionals.